Overview

Sensor Processing Graphs (SPG) enables running custom GPU code as post-processing passes on RTX render outputs (Arbitrary Output Variables, or AOVs). You write a CUDA kernel or a Slang shader, describe its launch configuration in a Lua script, declare its interface in USD, and wire it into a RenderProduct. All computation stays on the GPU; there is no CPU-side data transfer in the processing pipeline.

This guide assumes proficiency with CUDA kernel programming or Slang Shader programming and basic familiarity with USD (Universal Scene Description).

How SPG Works

Every SPG node is defined by three files:

File	Role
.cu or .slang	GPU source. The transform you write, in CUDA or in Slang.
.cu.lua or .slang.lua	Lua launch script. Named after the GPU source. Tells SPG how to validate the inputs, what the outputs have to be, and how to launch the GPU code.
.usda	USD shader definition. Declares the SPG node with its inputs and outputs in USD and references the GPU source. Wired into the render graph.

digraph { rankdir=TB bgcolor="transparent" node [shape=box style="rounded,filled" fontname="Helvetica" fontsize=11 penwidth=0] edge [color="#9aa0a6" penwidth=1.2 arrowsize=0.7] src [label="GPU source\n.cu / .slang" fillcolor="#76b900" fontcolor="#1b1b1b"] lua [label="launch script\n.lua" fillcolor="#76b900" fontcolor="#1b1b1b"] usda [label="shader definition\n.usda" fillcolor="#76b900" fontcolor="#1b1b1b"] spg [label="SPG" fillcolor="#3b4252" fontcolor="white"] out [label="Output AOV" fillcolor="#4c566a" fontcolor="white"] { rank=same; src lua usda } src -> spg lua -> spg usda -> spg spg -> out }

Finding Your Way

The pages after this one come in four kinds:

Section	Purpose
Your First Node	Walkthrough. Builds one node from nothing, in order, once.
How-To	One page per task. Find the one you need. Not written to be read in sequence.
Reference	Lookup. Every Lua function and every USD attribute you can author.
Troubleshooting	Symptoms. Starts from what you observed rather than from what you meant to do.